Some people use the helper all day. Some use it once a week. Give the all-day people their own helper, and let the once-a-week people pay a little each time.
Most teams have a few heavy Copilot users and many light ones. Giving everyone a licence wastes money on the light users, and giving nobody one frustrates the heavy users. Use the usage reports to sort people into groups, then check every quarter whether anyone has moved.
Segment users by observed demand: heavy (licence), light (Copilot Chat plus metered agents within a spending policy), and occasional (Copilot Chat only). Use active days and agent turns from usage reports, apply the break-even as a threshold, and rebalance quarterly.
Why not everyone, and why not nobody
The two easy answers to "who gets Copilot?" are everyone and nobody. Everyone is simple to run but pays for seats that sit idle. Nobody saves money but leaves heavy users doing work by hand that Copilot could help with. Most of the value I've seen is in between, where a team is split by how people actually work.
Three groups
I sort people into three groups. The boundaries are my assumptions and should move with your own data.
| Group | Typical pattern | What they get |
|---|---|---|
| Heavy | Uses Copilot on most working days, across several apps | Microsoft 365 Copilot licence |
| Light | Uses company agents or Copilot Chat a few times a week | Copilot Chat plus metered agent use under a spending policy |
| Occasional | Uses Copilot Chat now and then | Copilot Chat only |
Copilot Chat is included with eligible Microsoft 365 subscriptions, according to Microsoft, so the occasional group costs nothing extra. The light group's cost is whatever their agent use comes to in Copilot Credits. The heavy group's cost is the licence.
Where the line falls
The line between light and heavy comes from the break-even. On agent use alone, with the UK list price and an illustrative exchange rate, it's around 250 grounded agent turns a month. Anyone near that, or anyone who'd use Copilot inside Word, Excel, Outlook and Teams most days, goes in the heavy group.
A worked example
These numbers are illustrative.
A 30-person finance team has three months of usage data. Eight people use Copilot on 15 or more days a month. Twelve use a company policy agent about 40 times a month each. Ten barely use anything.
| Group | People | Monthly cost |
|---|---|---|
| Heavy, licensed (8 × £23.10) | 8 | £184.80 |
| Light, metered (12 × 40 turns × 12 credits × $0.01, at $1 = £0.77) | 12 | about £44 |
| Occasional, Copilot Chat only | 10 | £0 |
| Total | 30 | about £229 |
Licensing all 30 would cost £693 a month at list price. Licensing none would leave eight heavy users without Copilot in their apps. The split costs about a third of licensing everyone, and the people who'd notice a licence have one.
Moving people between groups
People change groups. A new project, a new role or a new agent can move someone from light to heavy in a month. I'd review the split quarterly using the Copilot usage report for licensed users and the Copilot Chat usage report for everyone else, as in Reading the Copilot usage report for licence decisions. Weekly reallocation suits metered allowance. Licences suit a slower rhythm, because a licence is bought on an annual subscription.
Being open about it
A split can feel like a ranking of people. I'd publish the rule, base it on use rather than seniority, and let anyone ask to move groups with a sentence about what they'd use it for. The request itself is useful data.
What I'm still checking
I haven't checked how Microsoft's annual subscription terms treat reassigning a licence mid-term between people. My understanding is that a licence can be moved from one user to another within the tenant, but the number of licences is fixed for the term. I'd confirm that against your own agreement before planning a quarterly rebalance around it.
Sources
- Microsoft UK, Microsoft 365 Copilot plans and pricing for enterprise, accessed 11 October 2026.
- Microsoft Learn, Billing rates and management, Copilot Studio, accessed 11 October 2026.
- Microsoft Learn, Microsoft Copilot usage report, accessed 11 October 2026.